March 2, 2017 | Written by: Cynthya Peranandam
Categorized: Artificial Intelligence
Share this post:
Significant advances in artificial intelligence over the past few years have broadened AI’s reach into industries such as healthcare, finance and even retail.
Businesses and consumers alike are benefiting from the rise of big data and the growth of AI techniques like deep learning and natural language processing. But we’re still only scratching the surface of what is possible with AI, and the full impact of the technology may be years away.
In the near-future, however, AI advances will give rise to increasingly powerful applications like personal assistants with more robust utility in the workplace and in our personal lives. These assistants could provide personalized information, help us make more informed decisions, and perhaps even provide physical assistance.
AI will also help us push the boundaries of creativity: It has already helped to write pop ballads and make movie trailers, but we may also see AI become a true partner that augments and elevates our creativity.
According to experts in a new report from IBM, making this next big leap to more pervasive AI will involve many factors, including the confluence of several forces.
First, AI solutions today require large amounts of public and proprietary data to train them to perform even the simplest of tasks, like recognizing an object. But diverse and industry-specific data isn’t always plentiful. In the future, AI experts believe smaller datasets will also enable AI innovation. This means AI systems could make decisions and provide results with less training.
And while there are many ways to “teach” AI to augment humans, the ultimate goal is to build advanced reasoning systems that require minimal training, especially for research-intensive fields such as medicine or life sciences. Experts say that while completely unsupervised learning might be a far-fetched goal, we’re getting closer to it as scientists find success with new, less labor-intensive teaching technique.
One such technique is transfer learning, which allows AI engineers to apply a trained model to completely new types of problems with little additional training. For example, applying the same model to solve business problems across different departments of a company. These less supervised techniques allow a machine’s deep network to absorb information from the world without a lot of hand holding or manual inputs.
Next, as we hit the limits of Moore’s Law, the power and efficiency crunch that slows the performance of AI applications is becoming more evident. The computational power to train and run these systems will greatly benefit from hardware innovation, including neuromorphic chips or even quantum computers powerful enough to process diverse information types simultaneously. While it is still early for some of these technologies, they promise exponential leaps in performance over today’s classical computers, potentially benefitting AI applications.
Finally, AI experts are quick to point out that getting to a state of pervasive AI will require more than technical breakthroughs. Building trust in AI systems through transparency will be key. This will mean general education on the benefits and pitfalls of AI, as well as opening up the guts of how AI decisions are made.
The process will also require researchers, architects and developers to create and share AI applications and techniques based on best practices for interoperability and seamless integration. And practitioners will need to be concerned about the ethics of AI programming and decision making. Finally, continued collaboration between academia, government and industry will be essential to build a common infrastructure and foster co-creation to accelerate AI advances.